Microsoft’s AI chatbot Tay was withdrawn 16 hours after its launch. It was using filthy language and puked venom. Facebook launched Galactica trained on 48 million papers, textbooks, lecture notes, scientific information and encyclopedias in September, 2022. The aim was to better organise scientific knowledge. Days after its launch, Galactica was withdrawn as it spewed suspicious and inaccurate information.

There should be transparency about the data the bot has been trained on, about the algorithms used. There should be accountability. The reference material must be publicly available. Anyone must be able to assess the quality and accuracy.